We present the results from the latest version of the GAlaxy Evolution and Assembly (GAEA) theoretical model of galaxy formation coupled with merger trees extracted from the Planck Millennium Simulation (PMS). With respect to the Millennium Simulation, typically adopted in our previous work, the PMS provides a better mass resolution (∼108 h-1 M⊙), a larger volume (8003 Mpc3), and assumes cosmological parameters consistent with latest results from the Planck mission. The model includes, at the same time, a treatment for the partition of cold gas into atomic and molecular (H2) components; a better treatment for environmental processes acting on satellite galaxies; an updated modelling of cold gas accretion on supermassive black holes, leading to the phenomenon of active galactic nuclei (AGN) and relative feedback on the host galaxy. We compare GAEA predictions based on the PMS, with model realizations based on other simulations in the Millennium Suite at different resolution, showing that the new model provides a remarkable consistency for most statistical properties of galaxy populations. We interpret this as being due to the interplay between AGN feedback and H2-based SFR (both acting as regulators of the cold gas content of model galaxies), as model versions considering only one of the two mechanisms do not show the same level of consistency. We then compare model predictions with available data for the galaxy two-point correlation function (2pCF) in the redshift range 0 < z ≤ 3. We show that GAEA runs correctly recover the main dependences of the 2pCF as a function of stellar mass (M∗), star formation activity, HI-content, and redshift for M∗ > 109 M⊙ galaxies. These results suggest that our model correctly captures both the distribution of galaxy populations in the large-scale structure and the interplay between the main physical processes regulating their baryonic content, both for central and satellite galaxies. The model predicts a small redshift evolution of the clustering amplitude that results in an overprediction of z ∼ 3 clustering strength with respect to the available estimates, but is still consistent with data within 1σuncertainties.